Press n or j to go to the next uncovered block, b, p or k for the previous block.
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 | 2x 2x 2x 2x 2x 2x 2x 2x 2x 6x 6x 6x 2x 16x 2x 2x 2x 2x 14x 14x 14x 16x 1x 1x 1x 1x 1x 1x 13x 13x 13x 15x 11x 11x 16x 2x 2x 2x 13x 13x 13x 13x 16x 13x 13x 16x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 31x 13x 13x 13x 13x 13x 13x 13x 13x 13x 13x 16x 16x 16x 16x 16x 16x 13x 13x 13x 13x 13x 13x 13x 16x 16x 16x 16x 16x 16x 16x 51x 51x 51x 3x 3x 3x 3x 3x 3x 51x 2x 2x 2x 2x 2x 2x 2x 4x 4x 2x 2x 2x 2x 2x 2x 2x 1x 1x 1x 1x 2x 2x 2x 2x 2x 2x 2x 2x 2x | /**
* The pairix index — `.px2` — and the two things it is actually for here.
*
* ## It is not a `.tbi`, and it says so in its first eight bytes
*
* pairix is a fork of tabix and the family resemblance is close enough to be a
* trap. The body is the same: a bin index and a linear index per reference,
* chunks as pairs of virtual offsets. The *header* is not, and it does not
* pretend to be — the magic is `PX2.004\1` rather than `TBI\1`, so
* `annotations/tabix.ts`'s `parseTabix` rejects a `.px2` outright instead of
* misreading one. That refusal is the good case and this module exists beside
* it rather than inside it for exactly that reason.
*
* What differs, verified against pairix's `index.c`:
*
* - **magic** is 8 bytes, and three versions are in circulation (`PX2.004\1`,
* `PX2.003\1`, `PX2.002\1`);
* - a **`linecount`** follows `n_ref` — 8 bytes in `.004`, 4 in the two older
* ones. This is the field that makes an import's cost knowable before it
* starts;
* - the **config block** is a raw dump of a C struct carrying *two* column
* triples (`sc/bc/ec` and `sc2/bc2/ec2`) for the two dimensions, plus a
* delimiter, a region-split character, a meta character and a line-skip;
* - a **reference name is a chromosome pair**, joined by the region-split
* character the header carries — `chr1|chr2`, not `chr1`.
*
* The config block is written with `sizeof(ti_conf_t)`, which means **compiler
* padding is part of the on-disk layout**. Summing the field widths gives 36
* bytes and a plausible-looking answer; a build that pads to 40 gives another.
* Rather than pick one and be silently wrong on half the files in the world,
* {@link parsePx2} tries each candidate and keeps the one whose *successor*
* fields make sense — the name block length has to be positive, has to fit, and
* has to split into exactly `n_ref` NUL-terminated names. A layout that
* satisfies all three is the layout.
*
* ## What is read, and what is deliberately not
*
* Only two things, because only two are wanted:
*
* 1. **`linecount`** — the exact number of records, before a byte of data is
* read. §12's "refused from the index alone" and §18.6's pre-flight estimate
* both want precisely this, and no other Hi-C container hands it over.
* 2. **One byte span per chromosome pair**, as the union of that key's chunks.
*
* The bin index is walked to compute (2) and then discarded, exactly as
* `parseTabix` discards it — and here the reason is stronger rather than
* weaker. An import bins *every record in scope* at a chosen resolution; there
* is no window to narrow to, so a finer index would be carried and never asked.
* What the spans buy is scope: binning four contigs of a metagenome without
* reading the other two thousand.
*/
import { BGZF_MAX_BLOCK, blockSize, inflateBgzf } from '../annotations/tabix'
import { ViewerError } from '../errors'
import type { ByteSource } from '../types'
/** The magics pairix has written, newest first; the flag is a 64-bit linecount. */
const MAGICS: Array<{ magic: string; wideLinecount: boolean }> = [
{ magic: 'PX2.004<', wideLinecount: true },
{ magic: 'PX2.003<', wideLinecount: false },
{ magic: 'PX2.002<', wideLinecount: false },
]
/**
* Candidate `sizeof(ti_conf_t)` values, in the order they are tried.
*
* 36 is the sum of the fields with natural alignment; 40 is what a build that
* rounds the struct up to its largest member's multiple produces. Both are
* validated rather than assumed — see the module docblock.
*/
const CONF_SIZES = [36, 40, 44]
/** Where one chromosome pair's records live, as compressed byte offsets. */
export interface PairSpan {
/** First BGZF block holding a record for this pair. */
from: number
/** Byte offset of the last block holding one; the block extends past it. */
to: number
}
export interface Px2Index {
/** Chromosome-pair keys in index order, as written (`chr1|chr2`). */
keys: string[]
/** Byte span per key. */
spans: Map<string, PairSpan>
/** Exact record count, from the header. */
linecount: number
/** The character joining the two chromosomes in a key. */
splitChar: string
/** The comment character records are skipped by. */
metaChar: string
/** Header lines skipped when the index was built. */
lineSkip: number
}
/** Split a `chr1|chr2` key into its two chromosomes. */
export function splitKey(key: string, splitChar: string): [string, string] {
const at = key.indexOf(splitChar)
return at < 0 ? [key, key] : [key.slice(0, at), key.slice(at + splitChar.length)]
}
/**
* Parse a `.px2`.
*
* The index is itself BGZF, so it is inflated whole first — safe for the same
* reason a `.tbi` is: it holds one entry per populated bin, not one per record.
*
* @throws ViewerError `format` when the magic is not pairix's, or when no
* candidate struct layout produces a coherent name block.
*/
export function parsePx2(inflated: Uint8Array): Px2Index {
if (inflated.length < 20) {
throw new ViewerError('format', 'This .px2 index is too short to hold a header.', {
bytes: inflated.length,
})
}
const view = new DataView(inflated.buffer, inflated.byteOffset, inflated.byteLength)
const magic = String.fromCharCode(...inflated.subarray(0, 8))
const known = MAGICS.find((entry) => entry.magic === magic)
if (known === undefined) {
throw new ViewerError(
'format',
'This is not a pairix index: the magic bytes are not PX2. A `.tbi` cannot be used ' +
'for a pairs file — build one with `pairix <file>.pairs.gz`.',
{ magic: magic.replace(/[^\x20-\x7e]/g, '?') },
)
}
const nRef = view.getInt32(8, true)
let at = 12
let linecount: number
if (known.wideLinecount) {
linecount = Number(view.getBigUint64(at, true))
at += 8
} else {
linecount = view.getUint32(at, true)
at += 4
}
const layout = resolveConfLayout(view, inflated, at, nRef)
const { splitChar, metaChar, lineSkip, namesAt, nameBytes } = layout
const keys: string[] = []
const raw = new TextDecoder().decode(inflated.subarray(namesAt, namesAt + nameBytes))
for (const name of raw.split('\0')) if (name !== '') keys.push(name)
let cursor = namesAt + nameBytes
const spans = new Map<string, PairSpan>()
for (let ref = 0; ref < nRef; ref++) {
if (cursor + 4 > inflated.length) break
const nBin = view.getInt32(cursor, true)
cursor += 4
let from = Number.POSITIVE_INFINITY
let to = Number.NEGATIVE_INFINITY
for (let b = 0; b < nBin; b++) {
if (cursor + 8 > inflated.length) break
// The bin number is not kept: every chunk of every bin belongs to this
// key, and the span this module serves is their union.
cursor += 4
const nChunk = view.getInt32(cursor, true)
cursor += 4
for (let c = 0; c < nChunk; c++) {
if (cursor + 16 > inflated.length) break
// A virtual offset is `(block start << 16) | offset within block`, and
// the block start alone passes 2^32 on any file over four gigabytes —
// which a pairs file very much is. BigInt, not a `getUint32` pair.
const begin = Number(view.getBigUint64(cursor, true) >> 16n)
const end = Number(view.getBigUint64(cursor + 8, true) >> 16n)
cursor += 16
if (begin < from) from = begin
if (end > to) to = end
}
}
if (cursor + 4 > inflated.length) break
const nIntv = view.getInt32(cursor, true)
cursor += 4
for (let i = 0; i < nIntv; i++) {
if (cursor + 8 > inflated.length) break
const offset = Number(view.getBigUint64(cursor, true) >> 16n)
cursor += 8
// The linear index only ever *lowers* the start.
if (offset > 0 && offset < from) from = offset
}
if (Number.isFinite(from) && Number.isFinite(to) && ref < keys.length) {
spans.set(keys[ref], { from, to })
}
}
return { keys, spans, linecount, splitChar, metaChar, lineSkip }
}
interface ConfLayout {
splitChar: string
metaChar: string
lineSkip: number
namesAt: number
nameBytes: number
}
/**
* Pick the struct layout whose successor fields are coherent.
*
* The validation is the point: a wrong candidate does not produce a slightly
* wrong answer, it produces a `nameBytes` that is negative, enormous, or runs
* off the end of the buffer — all three of which are checked, along with the
* name block actually holding `n_ref` names.
*/
function resolveConfLayout(
view: DataView,
bytes: Uint8Array,
confAt: number,
nRef: number,
): ConfLayout {
const tried: number[] = []
for (const size of CONF_SIZES) {
const lengthAt = confAt + size
if (lengthAt + 4 > bytes.length) continue
const nameBytes = view.getInt32(lengthAt, true)
const namesAt = lengthAt + 4
tried.push(nameBytes)
if (nameBytes <= 0 || namesAt + nameBytes > bytes.length) continue
const names = new TextDecoder()
.decode(bytes.subarray(namesAt, namesAt + nameBytes))
.split('\0')
.filter((name) => name !== '')
if (names.length !== nRef) continue
// The three single-byte fields sit immediately after the seven uint32s.
const charsAt = confAt + 28
return {
splitChar: String.fromCharCode(bytes[charsAt + 1] || 0x7c),
metaChar: String.fromCharCode(bytes[charsAt + 2] || 0x23),
lineSkip: view.getInt32(confAt + 32, true),
namesAt,
nameBytes,
}
}
throw new ViewerError(
'format',
'This .px2 index has a header this reader cannot lay out: no candidate config size ' +
`produced a coherent name block for ${nRef} references. Rebuild it with ` +
'`pairix <file>.pairs.gz`, or open the file without an index.',
{ nRef, candidates: CONF_SIZES, sawNameBytes: tried },
)
}
/**
* Load and parse the `.px2` sitting beside a pairs file, or `null`.
*
* `null` rather than a throw for an *absent* index, because an unindexed file
* is a supported input and not an error — the import reads it whole and says
* so. A present-but-broken index does throw, since silently ignoring one is how
* a reader ends up scanning eleven gigabytes with no explanation.
*/
export async function loadPx2(open: () => Promise<ByteSource | null>): Promise<Px2Index | null> {
const source = await open()
if (source === null) return null
try {
const bytes = source.readBytes(0, source.size)
return parsePx2(await inflateBgzf(bytes))
} finally {
source.close()
}
}
/** A byte range to read, merged across the keys asked for. */
export interface ByteSpan {
from: number
to: number
}
/**
* The byte spans covering `wanted`, merged where they touch.
*
* The end of each span is made **exact** rather than padded: `to` names the
* start of the last block holding a record, and that block's own length is in
* its header, so one 12-byte read turns a 64 KiB over-ask into a precise
* boundary. That matters more here than it does for a tabix span, because the
* result is fed to `DecompressionStream('gzip')` as a whole — and a truncated
* final member is a decode error rather than a few wasted bytes.
*/
export function spansFor(
index: Px2Index,
wanted: readonly string[],
source: ByteSource,
): ByteSpan[] {
const spans: ByteSpan[] = []
for (const key of wanted) {
const span = index.spans.get(key)
if (span === undefined) continue
spans.push({ from: span.from, to: endOfBlockAt(source, span.to) })
}
spans.sort((a, b) => a.from - b.from)
const merged: ByteSpan[] = []
for (const span of spans) {
const last = merged[merged.length - 1]
// A gap smaller than a source chunk costs more to seek over than to read
// across — the same argument `CoolerStore.pixelBandsOf` makes about pixel
// bands, and `tabix.ts` about sequence spans.
if (last !== undefined && span.from <= last.to + (1 << 20)) {
last.to = Math.max(last.to, span.to)
} else {
merged.push({ ...span })
}
}
return merged
}
/** One past the end of the BGZF block beginning at `at`. */
function endOfBlockAt(source: ByteSource, at: number): number {
if (at >= source.size) return source.size
const header = source.readBytes(at, Math.min(BGZF_MAX_BLOCK, source.size - at))
const size = blockSize(header)
return size === null ? Math.min(source.size, at + BGZF_MAX_BLOCK) : Math.min(source.size, at + size)
}
|